12 July 2018

Content

  • Introduction to metabolomics
  • Preprocessing of LC-MS data in Bioconductor
  • Normalization
  • Annotation/identification

Metabolite? Metabolism?

Metabolite? Metabolism?

Metabolite? Metabolism?

Metabolite? Metabolism?

Metabolite? Metabolism?

Metabolite? Metabolism?

Metabolite? Metabolism?

  • Metabolites: intermediates and products of cellular processes.

Metabolomics?

  • Large-scale study of small molecules (metabolites) in a system (cell, tissue, organism).
  • Comparison of the different -omes:
  • Genome: what can happen.
  • Transcriptome: what appears to be happening.
  • Proteome: what makes it happen.
  • Metabolome: what actually happened.
  • Metabolome influenced by genetic and environmental factors.

How can we measure metabolites?

  • Nuclear Magnetic Resonance (NMR) - not covered here.
  • Mass spectrometry (MS)-based metabolomics.
  • Targeted/untargeted metabolomics:
  • targeted: quantitative measurement of selected metabolites.
  • untargeted: semi-quantitative measurement of all metabolites (detectable with the setup) in a sample.

Mass Spectrometry (MS)

Mass Spectrometry (MS)

  • Problem: unable to distinguish between metabolites with the same mass-to-charge ratio (m/z).
  • Solution: separate metabolites prior to MS by another property.

Liquid chromatography

  • Sample is dissolved in a fluid (mobile phase).

Liquid chromatography

  • Sample is dissolved in a fluid (mobile phase).

  • Mobile phase carries analytes through a column with a stationary phase.

Liquid chromatography

  • Sample is dissolved in a fluid (mobile phase).

  • Mobile phase carries analytes through a column with a stationary phase.

  • Separation based on affinity for the column’s stationary phase.

Liquid chromatography

  • Sample is dissolved in a fluid (mobile phase).

  • Mobile phase carries analytes through a column with a stationary phase.

  • Separation based on affinity for the column’s stationary phase.

  • HILIC (hyrophilic liquid interaction chromatography):

    • Hydrophilic, polar stationary phase.

    • Analytes solved in mobile phase.

    • Analytes separated by polarity: low polarity elute first, high polarity later.

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

Liquid Chromatography Mass Spectrometry (LC-MS)

LC-MS data has thus 3 dimensions:

  • intensity, m/z, retention time tuples.
  • retention time not constant between samples.
  • m/z not identical between samples.

LC-MS data preprocessing

  • Chromatographic peak detection
  • Alignment
  • Correspondence

Chromatographic peak detection

  • Aim: identify chromatographic peaks in the data.

Chromatographic peak detection

  • Aim: identify chromatographic peaks in the data.

Chromatographic peak detection

  • Aim: identify chromatographic peaks in the data.
  • allow different rt-widths
  • allow some scattering on m/z
  • identify peak boundaries

Chromatographic peak detection

  • centWave [Tautenhahn et al. BMC Bioinformatics, 2008]:
  • Step 1: identify regions of interest.

Chromatographic peak detection

  • Step 2: peak detection using continuous wavelet transform.
  • Allows detection of peaks with different widths.

Chromatographic peak detection

  • After reading the data with readMSData (MSnbase package):
  • xcms: findChromPeaks function, passing settings along with an algorithm-specific parameter object.
cwp <- CentWaveParam(peakwidth = c(2, 10), snthresh = 5)
data <- findChromPeaks(data, param = cwp)
head(chromPeaks(data), n = 3)
##            mz    mzmin    mzmax     rt rtmin  rtmax     into     intb
## [1,] 114.0907 114.0899 114.0929  1.954 0.280  3.907 1559.829 1555.923
## [2,] 114.0913 114.0884 114.0929  5.860 4.465  8.650 1890.221 1885.757
## [3,] 114.0914 114.0899 114.0929 10.882 8.650 13.114 1950.953 1946.210
##          maxo  sn sample is_filled
## [1,] 584.9510 584      1         0
## [2,] 601.8881 601      1         0
## [3,] 691.9580 691      1         0

Alignment

  • Chromatography subject to (random and systematic) noise.
  • Same analyte may elute at different time in different runs.

  • Shifts are LC-setup dependent, seem to be also analyte dependent.

Alignment

  • How strong the shifts are depends on the LC-setup.
  • Many algorithms available [Smith et al. Brief Bioinformatics 2013]
  • Main assumption: analytes elute in the same order.
  • xcms: adjustRtime function with PeakDensityParam [Smith et al. Anal. chem. 2006] or ObiwarpParam [Prince et al. Anal. chem. 2006].

Correspondence

  • Aim: group peaks across samples, assuming they represent the same ion.
  • Depends on proper alignment.

Correspondence

  • Aim: group peaks across samples, assuming they represent the same ion.
  • Depends on proper alignment.

Correspondence

  • Aim: group peaks across samples, assuming they represent the same ion.
  • Depends on proper alignment.

Correspondence

  • xcms: groupChromPeaks with NearestPeaksParam [Katajamaa et al. Bioinformatics 2006] and PeakDensityParam [Smith et al. Anal. chem. 2006].
  • peak density approach (for a given m/z slice):

Correspondence

  • xcms: groupChromPeaks with NearestPeaksParam [Katajamaa et al. Bioinformatics 2006] and PeakDensityParam [Smith et al. Anal. chem. 2006].

  • peak density approach (for a given m/z slice):

Normalization

  • Batch effects.
  • LC-dependent effects: seem to affect each metabolite in a different way.

Normalization

  • QC controls required for proper noise estimation and adjustment.
  • Commonly used: pooled samples measured repeatedly.
  • MS runs usually not very expensive, running replicates, QC controls etc not a problem.
  • Popular methods:
    • RUV [De Livera et al. Anal. Chem. 2015]
    • linear models [Wehrens et al. Metabolomics 2016]
    • linear and higher order models [Brunius et al. Metabolomics 2016].

Identification

  • Features have to be annotated to metabolites.
## DataFrame with 4 rows and 4 columns
##                  mzmed            rtmed           POOL_1           POOL_2
##              <numeric>        <numeric>        <numeric>        <numeric>
## FT001 105.041814839707 167.961095453642 229.490739260736 3093.75184315684
## FT002 105.041653033614 157.083057856508 4762.39872227772 6601.45091358641
## FT003 105.069636149683 31.8108067962868 699.723986763235 1033.23232267732
## FT004  105.11027064078 63.7513630255991 20211.2633706294 15839.5504368189

  • Match compounds based on features’ m/z.

  • Will result in an 1:n mapping.

Identification based on m/z matching

  • Databases containing masses of (synthetic and biological) compounds.
  • m/z is not the mass.
  • The mass of the charged molecule has to be considered too: [M+H]+ adduct: mass of metabolite M + mass of hydrogen.
  • We can have different adducts from the same compound in a sample: [M+H]+, [M+Na]+, …

Improved identification

Identify compounds based on m/z and:

  • retention time: requires lab-internal database with approximate retention times for specific compounds (manually defined).
  • MS2 spectrum:
    • Needs MS2 data of the compounds (e.g. SWATH data).
    • Only works if reference spectrum available in database.
  • Up and coming: CompoundDb package (similar to ensembldb and alike).

Afternoon metabolomics lab

  • LC-MS data handling (MSnbase).
  • LC-MS data preprocessing using xcms.